Wind turbine gearbox fault diagnosis based on an improved supervised autoencoder using vibration and motor current signals
Shuai Yang, Yu Wang, Chuan Li
Abstract
Abstract It is an important research topic to learn features automatically from raw signals of wind turbines. Although vibration signals have been widely used for fault diagnosis, the inevitable disturbances and measurement noise make it difficult for vibration signals to completely reflect the fault conditions. In order to capture more sensitive features and generate accurate fault diagnosis results, a multi-source data fusion method based on an autoencoder (AE) is proposed in this paper, where the vibration and motor current signals are adopted. Compared with standard AE, the proposed method applies label information to the training process of AE, which is achieved by adding a new supervised penalty term to the hidden layer of AE. In this way, the learned features are supposed to be close to the class features within, and separate from the inter-class features; thus, the separability of the learned features is enhanced. The proposed method is used to merge the vibration signal and motor current signal of the wind turbine gearbox, and the results show that the fused data generated by the proposed method achieve a higher and more robust performance.